RNA interference (RNAi) is a small RNA-guided gene silencing process within living cells. The RNAi technique is widely used in both biological research and clinical applications because it has the ability to knockdown essentially any gene of interest. However, one major unresolved issue in RNAi studies is non-specific gene silencing. It is well known that, in addition to the single intended gene target, many unintended targets are also simultaneously silenced. Thus, there is an urgent need to develop new methods for improving RNAi specificity. RNAi specificity and potency are determined by the small interfering RNA (siRNA) in the silencing complex. We propose the hypothesis that gene silencing specificity can be significantly improved with rational siRNA design. We have previously developed a machine learning algorithm for the design of widely-distributed commercial siRNAs. Based on this commercial success, we propose to further develop a new method for genome-wide design of next-generation siRNAs with significantly reduced off-target effects. The new design method will lay a solid foundation for further commercial development of siRNA products that can be used in a variety of RNAi-based applications.